Abstract
This study introduces the "p-index," a data-driven metric adapted from the academic h-index, to evaluate the walkability and bikeability of urban roads using crowdsourced GPS tracks. By analyzing data from Wuhan, China, the research demonstrates how this metric filters out anomalies—such as one-off sporting events—to reveal genuine, consistent human behavior patterns across different times of day.
Introduction
Urban planners traditionally rely on physical infrastructure audits (e.g., sidewalk width) or static surveys to assess non-motorized transport (NMT) systems. However, these methods often fail to capture actual human behavior and ground-truth usage. While GPS data from fitness apps offers a potential solution, raw counts are highly susceptible to noise from isolated events like marathons or races, which can artificially inflate a road's perceived popularity. This paper addresses the need for a robust metric that identifies genuine, repeated road usage by multiple individuals.
Methodology
The core innovation of this research is the p-index (popularity index), heavily inspired by the bibliometric h-index. A road segment is assigned a p-index of h if h distinct individuals have used it as a route at least h times. This dual requirement for breadth (many unique users) and depth (frequent, repeated use) effectively isolates consistent daily usage from sporadic activity spikes.
Data Processing
The study utilized two primary crowdsourced datasets from Wuhan, China: Edooon (walking/running) and XingZhe (cycling). The data pipeline involved:
- Cleaning: Applying speed filters (<25 km/h for walking, <35 km/h for cycling) and removing tracks with significant GPS gaps.
- Map Matching: Snapping raw coordinates to the urban road graph using a Hidden Markov Model (HMM).
- Temporal Segmentation: Dividing the 24-hour cycle into five distinct periods based on observed activity peaks to capture shifting urban dynamics.
Temporal and Spatial Dynamics
Human activity is not uniform throughout the day. The researchers identified five distinct temporal periods—late-night, pre-work, on-work, off-work, and night activities—based on observed peaks in the crowdsourced data.
As illustrated in Figure 2, both walking and cycling exhibit similar temporal trends, with significant peaks during morning commutes (7:00–9:00) and evening leisure or exercise hours (19:00–21:00). These peaks informed the segmentation of the p-index analysis to better understand how road popularity fluctuates.
Key Results and Insights
The p-index successfully mapped the NMT landscape of Wuhan, revealing several key behavioral insights:
- Mode-Specific Geographies: Walkability is highly localized and clustered around residential zones, whereas bikeability is characterized by long, continuous corridors along major urban arteries.
- Preference for "Blue Space": The data highlights a strong human preference for scenic, water-adjacent routes. Roads near the Yangtze River and local lakes (South Lake, East Lake) consistently achieved the highest p-index scores.
- Temporal Shifts: Popularity is dynamic. For instance, late-night walking is almost exclusively concentrated on bridges crossing the Yangtze River, while evening workouts shrink back to specific high-quality corridors.
Figure 6 demonstrates the spatial distribution of these popularity scores. While certain corridors remain popular throughout the day, industrial and logistical zones (like Qingshan and Dongxihu) consistently showed low p-index scores, marking them as hostile environments for non-motorized transport.
Conclusion and Future Work
The p-index provides a scalable, lightweight heuristic for urban analysis that effectively filters noise from crowdsourced mobility data. While the study is limited by the demographic bias inherent in fitness app users, it offers a powerful tool for identifying high-quality NMT corridors. The authors propose that these insights could be integrated into future navigation systems to suggest "human-friendly" routes based on historical popularity rather than just the shortest physical distance.
Summary updated